When AI gets it wrong in front of a customer: what to say
It will happen. What you say in the next hour determines whether it is a small embarrassment or a lasting doubt about your company.
Eventually a wrong answer reaches a customer. A delivery date that was never possible, a figure that does not match, a technical claim that is not true. The system worked as designed, the reviewer was busy, and it went out.
The technical response is covered elsewhere in this blog. This is about the hour afterwards, which most companies handle badly in one specific way.
The mistake: leading with the tool
« That came from our AI system » feels like an explanation. To a customer it is one of two things, and neither is good.
Either it sounds like blame-shifting to software you chose — which does not transfer responsibility anywhere a customer recognises. Or it raises a question they had not asked: what else here is machine-generated, and does anyone check it?
One wrong delivery date is an incident. A customer newly wondering whether your quotations are reviewed is a structural problem.
What to say instead
Correct it first, in one sentence, without context. The right date, the right figure, plainly. The customer needs the fact before they need the story.
Own it without a mechanism. « That was wrong and it should not have reached you. » No subject other than your company.
Say what changes, specifically. Not « we will review our processes », which everyone recognises as nothing. « Delivery dates now get confirmed against the production plan before a quotation goes out » is a sentence with a fact in it.
When to mention it at all
If they ask directly, answer honestly and immediately — evasion here is much worse than the original error. And if you use AI in something that visibly affects them, such as a website assistant, tell them before it goes wrong rather than after.
Otherwise, the customer does not care which of your internal tools produced the error. They care that it happened, that you fixed it, and that you can tell them why it will not happen again.
The underlying point
Your customer relationship is with your company, not with your software stack. The review step this blog has recommended since December exists precisely so that when something goes wrong, the answer is « we made a mistake » rather than « the machine did ».
The first is recoverable. The second invites a question you do not want asked.